ankanbhunia/Handwriting-Transformers

Implements transformer-based styled handwritten text synthesis using encoder-decoder attention to disentangle style and content while capturing global and local writing patterns through self-attention. Supports arbitrary text lengths and few-shot style transfer, generalizing to unseen words and writing styles. Trainable on custom datasets (IAM, CVL) with PyTorch; includes Hugging Face and Colab demos for inference on custom handwriting samples.

252 stars. No commits in the last 6 months.